The article introduces pkCSM, a novel computational method for predicting small-molecule pharmacokinetic (PK) and toxicity properties using graph-based signatures. This approach leverages molecular structure and chemical properties to develop predictive models for ADMET (absorption, distribution, metabolism, excretion, toxicity) properties. pkCSM performs as well as or better than existing methods and is available as a freely accessible web server (http://structure.bioc.cam.ac.uk/pkcsm), which allows users to rapidly evaluate PK and toxicity properties of compounds. The method uses graph-based signatures derived from molecular structures, incorporating both general molecular properties and distance-based graph features. These features are used to train machine learning models that predict various ADMET properties, including absorption, distribution, metabolism, excretion, and toxicity. The system includes 30 predictors, divided into five major classes, and has been validated on multiple datasets, showing statistically significant improvements in predictive performance compared to existing tools. pkCSM's ability to handle large datasets and its user-friendly web interface make it a valuable tool for medicinal chemists to evaluate and optimize compound properties early in the drug development process. The method has been tested on various datasets, including rat toxicity, Caco2 permeability, and P-glycoprotein inhibitors, demonstrating its effectiveness in predicting ADMET properties. The system also provides a platform for analyzing and optimizing pharmacokinetic and toxicity properties, helping to reduce the attrition rate in drug development by identifying compounds with favorable properties early on. The web server does not retain any submitted data, ensuring privacy and facilitating the rapid design, evaluation, and prioritization of compounds.The article introduces pkCSM, a novel computational method for predicting small-molecule pharmacokinetic (PK) and toxicity properties using graph-based signatures. This approach leverages molecular structure and chemical properties to develop predictive models for ADMET (absorption, distribution, metabolism, excretion, toxicity) properties. pkCSM performs as well as or better than existing methods and is available as a freely accessible web server (http://structure.bioc.cam.ac.uk/pkcsm), which allows users to rapidly evaluate PK and toxicity properties of compounds. The method uses graph-based signatures derived from molecular structures, incorporating both general molecular properties and distance-based graph features. These features are used to train machine learning models that predict various ADMET properties, including absorption, distribution, metabolism, excretion, and toxicity. The system includes 30 predictors, divided into five major classes, and has been validated on multiple datasets, showing statistically significant improvements in predictive performance compared to existing tools. pkCSM's ability to handle large datasets and its user-friendly web interface make it a valuable tool for medicinal chemists to evaluate and optimize compound properties early in the drug development process. The method has been tested on various datasets, including rat toxicity, Caco2 permeability, and P-glycoprotein inhibitors, demonstrating its effectiveness in predicting ADMET properties. The system also provides a platform for analyzing and optimizing pharmacokinetic and toxicity properties, helping to reduce the attrition rate in drug development by identifying compounds with favorable properties early on. The web server does not retain any submitted data, ensuring privacy and facilitating the rapid design, evaluation, and prioritization of compounds.